Lactate Correlation Paper: Nanostring Analysis of All Data
John Santiago
2022-03-23
Table of Contents
Section 1: Expression analysis and model generation using original 10 samples.
Figure 1: Barplot of correlation between biomarker expression and lactate levels
Figure 2: Heatmap of scaled biomarker expression data
Figure 3: Correlation between mean scaled biomarker expression and lactate
Figure 4: Multiple linear regression predictive model
Figure 5: Checking model conditions
Figure 6: Correlation between the models predicted lactate levels and donor metrics
Section 2: Incorporating additional samples into the predictive model.
Figure 7: Incorporating the additional samples into the established model
Figure 8: Correlation between the models predicted lactate levels for all samples and donor metrics
Figure 9: Expression levels of all biomarkers with lactate for all samples
Figure 10: Heatmap of scaled biomarker expression data for all samples
Figure 11: Correlation between mean scaled biomarker expression and lactate for all samples
Figure 12: PCA using scaled biomarker data for all samples
Figure 13: PCA using scaled biomarker data for the original 10 samples only
Figure 14: Multiple Linear Regression Predictive Model generated using all samples
Figure 15: Checking the new model conditions
Figure 16: Correlation between the new models predicted lactate levels for all samples and donor metrics
Section 3: Additional analyses.
Figure 17: Generating a model using genes that do not correlate with lactate.
Figure 18: Correlation between gene expression and 3 hour lactate levels for DBD livers.
Figure 19: Correlation between gene expression and 3 hour lactate levels for DBD and DCD livers.
Figure 20: Correlation of genes that correlate with lactate in DBD livers in DBD(NS), DCD(NS), and DCD(RNAseq).
Figure 21: Fold change between adequate and low functioning livers for genes that correlate with lactate in DBD livers in DBD(NS), DCD(NS), and DCD(RNAseq).
Figure 22: Expression levels for genes that correlate with lactate in DBD livers in DBD(NS), DCD(NS), and DCD(RNAseq).
Summary of Section 3.
The model using non-correlating genes is not as accurate as the model using correlating genes (Fig17).
DBD livers do have genes that correlate with lactate in the nanostring data (Fig18 and Fig20).
Combined data for DBD and DCD does not have correlating genes in the nanostring data (Fig19)
One of the DBD correlating genes (SERF2) has sufficient fold change (Fig21) and expression (Fig22) to act as biomarkers.
There are significant discrepencies observed between the nanostring and RNAseq data for several of these genes in correlation (fig20) and fold change (fig21) of DCD samples.
One of the genes (NRF1) correlates (fig20), has sufficient fold change (fig21) and expression in the nanostring data but not in the RNAseq data for the DCD samples.
Generating a model using the original biomarker data from DCD samples and incorporating fwit does mildly improve the model.
Section 1: Expression analysis and model generation using nanostring data.
Expression levels of putative biomarkers identified with RNAseq were validated using a custom nanstring panel and tested for conserved correlation with 3 hour lactate levels. Genes that met the biomarker criteria in both analyses were then used to generate a multiple linear regression predictive model.
Figure 1: The correlation of sufficiently expressed candidate biomarkers with 3 hour lactate levels. Genes needed to be above the negative control threshold in at least 8/10 samples to be considered. Dashed lines represent the 80% correlation threshold.
Figure 2: Heatmap with biomarker expression data scaled by gene/row and hierarchally clustered by sample/column. Functional assessment based on the 3 hour lactate levels is labeled above the columns. Data for the 7 biomarkers with sufficient expression and correlation with lactate were used from all 10 of the original liver samples.
Figure 3: Mean relative expression of the geneset correlated with 3 hour lactate levels. Light grey lines are the relative expression levels of each biomarker. The black line is the mean of the relative expression levels with error bars representing the variance. The dashed red line is the lactate level observed for each sample at 3 hours of perfusion. The Pearson Correlation coefficient and p-value for the correlation between mean relative expression and lactate are listed in the legend.
##
## Call:
## lm(formula = LacData ~ EPHX1 + TKT + GPX2 + JUN + CYP2B6 + GSTA1 +
## GSTA2, data = mlm.data)
##
## Residuals:
## FV2 LV2 LV3 FV3 LV1 FN1 FN2
## -1.9455673 -0.5191204 -0.8084125 -0.3786886 2.9397076 0.8948823 -0.2792483
## LN1 FN3 LN3
## -0.0008817 -0.0170392 0.1143681
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1.279e+00 2.340e+00 -0.546 0.640
## EPHX1 -2.580e-04 1.559e-03 -0.165 0.884
## TKT 1.314e-02 3.020e-02 0.435 0.706
## GPX2 -7.046e-04 1.914e-03 -0.368 0.748
## JUN 1.434e-02 3.160e-02 0.454 0.694
## CYP2B6 -5.606e-05 3.487e-04 -0.161 0.887
## GSTA1 6.208e-05 1.708e-04 0.364 0.751
## GSTA2 6.670e-04 1.108e-03 0.602 0.608
##
## Residual standard error: 2.682 on 2 degrees of freedom
## Multiple R-squared: 0.9723, Adjusted R-squared: 0.8752
## F-statistic: 10.02 on 7 and 2 DF, p-value: 0.09376Figure 4: Generating a Multiple Linear Regression Predictive Model and applying it to the observed expression levels. The model is generated by incorporating expression data from the 7 biomarker genes across the 10 original sample livers and calculating a functional relationship to the lactate levels observed in their corresponding sample. The resulting intercept and coefficients for individual gene expression levels can be used to predict lactate levels for a sample. The plot displays the relationship between the levels predicted by the model when gene expression levels observed in the original 10 sample livers are used and the actual observed levels for those livers. The dashed line represents a 1:1 relationship between obersved and predicted lactate levels.
Figure 5: Checking model conditions. These figures are used to check the quality of the model by testing specific statistical parameters that are required for reliable application.
Figure 6: The predicted lactate levels generated by the model were tested for correlation with the known donor metrics.
Section 2: Incorporating additional samples into the model.
In order to validate the capabilities of these genes to serve as biomarkers, expression was measured in 7 additional livers with the custom nanostring panel.
Figure 7: Incorporating the additional samples into the established model
Figure 8: Correlation between the models predicted lactate levels and donor metrics
Figure 9: Expression levels of the 23 putative biomarkers observed in all 17 samples correlated with 3 hour lactate levels. The black lines represent 80% correlation and the red represents 60% correlation
Figure 10: Heatmap with data scaled by gene/row and clustered by sample/column. Functional data and DCD/DBD status is labeled above the columns
Figure 11: Mean relative expression of the geneset using all samples correlated with 3 hour lactate levels. Light grey lines are the relative expression levels of each biomarker. The black line is the mean of the relative expression levels with error bars representing the variance. The dashed red line is the lactate level observed for each sample at 3 hours of perfusion. The Pearson Correlation coefficient and p-value for the correlation between mean relative expression and lactate are listed in the legend.
Figure 12: PCA using scaled data for all 17 samples
Figure 13: PCA using scaled data for the original 10 samples only
##
## Call:
## lm(formula = all.LacData ~ EPHX1 + TKT + GPX2 + JUN + CYP2B6 +
## GSTA1 + GSTA2, data = mlm.data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -4.7214 -1.9324 -0.6724 1.4903 9.1349
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.1045178 2.5151175 0.042 0.9678
## EPHX1 -0.0019640 0.0010002 -1.964 0.0812 .
## TKT 0.0267848 0.0103976 2.576 0.0299 *
## GPX2 0.0002049 0.0011610 0.176 0.8638
## JUN 0.0009719 0.0030291 0.321 0.7556
## CYP2B6 0.0001328 0.0002971 0.447 0.6653
## GSTA1 0.0002007 0.0001725 1.163 0.2747
## GSTA2 0.0010639 0.0006957 1.529 0.1605
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4.287 on 9 degrees of freedom
## Multiple R-squared: 0.8131, Adjusted R-squared: 0.6677
## F-statistic: 5.592 on 7 and 9 DF, p-value: 0.01012Figure 14: Generating a Multiple Linear Regression Predictive Model using all samples and applying it to the observed expression levels of all 17 samples
Figure 15: Checking the new model conditions
Figure 16: Correlation between the new models predicted lactate levels and donor metrics
Section 3: Additional analyses.
## MCL1 OAZ1 ATG16L1 ATG12 NDUFA13 TFAM BCL2
## 0.4920644 0.1541728 0.4983044 0.3187156 0.4602579 0.3698896 0.2063185##
## Call:
## lm(formula = LacData ~ gene1 + gene2 + gene3 + gene4 + gene5 +
## gene6 + gene7, data = mlm.data)
##
## Residuals:
## FV2 LV2 LV3 FV3 LV1 FN1 FN2 LN1 FN3 LN3
## -2.2358 -5.5515 -0.4721 0.9563 2.2442 -0.7144 3.6764 -2.0387 -0.8576 4.9934
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 1.879247 14.125759 0.133 0.906
## gene1 -0.006514 0.033210 -0.196 0.863
## gene2 -0.009600 0.006545 -1.467 0.280
## gene3 0.171600 0.215920 0.795 0.510
## gene4 0.039397 0.075945 0.519 0.656
## gene5 0.032455 0.039829 0.815 0.501
## gene6 0.034269 0.098919 0.346 0.762
## gene7 -0.320129 0.261545 -1.224 0.346
##
## Residual standard error: 6.551 on 2 degrees of freedom
## Multiple R-squared: 0.8345, Adjusted R-squared: 0.2552
## F-statistic: 1.441 on 7 and 2 DF, p-value: 0.4692Figure 17: Generating a model using genes that do not correlate with lactate. Uses a random set of 7 genes from the nanostring data that have sufficient expression and a positive correlation value less than .5.
Figure 18: The correlation of sufficiently expressed genes detected in the nanostring custom panel with 3 hour lactate levels in DBD livers. Genes needed to be above the negative control threshold in at least 8/10 samples to be considered. Dashed lines represent the 80% correlation threshold.
Figure 19: The correlation of sufficiently expressed genes detected in the nanostring custom panel with 3 hour lactate levels in DBD and DCD livers. Genes needed to be above the negative control threshold in at least 8/10 samples to be considered. Dashed lines represent the 80% correlation threshold.
Figure 20: Showing only those with at least 80% correlation in the DBD livers. Note: NRF1 was not used in the original analysis of DCD livers since there was a low correlation between the Lactade data and RNAseq data.
Figure 21: Showing the fold change between adequate and low functioning livers for those genes with at least 80% correlation. Note: NRF1 was not used in the original analysis of DCD livers since there was a low correlation between the Lactade data and RNAseq data.
Figure 22: Showing the expression leves for those genes with at least 80% correlation.
Figure 22: Showing the expression leves for those genes with at least 80% correlation.
Figure 22: Showing the expression leves for those genes with at least 80% correlation.
##
## Call:
## lm(formula = all.LacData ~ EPHX1 + TKT + GPX2 + JUN + CYP2B6 +
## GSTA1 + GSTA2 + fwit, data = mlm.data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -4.2035 -2.0985 -0.7066 1.4506 9.0308
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 1.491e+00 4.005e+00 0.372 0.719
## EPHX1 -1.919e-03 1.052e-03 -1.825 0.105
## TKT 2.902e-02 1.193e-02 2.433 0.041 *
## GPX2 -8.473e-05 1.369e-03 -0.062 0.952
## JUN -5.877e-04 4.645e-03 -0.127 0.902
## CYP2B6 1.518e-04 3.137e-04 0.484 0.642
## GSTA1 2.264e-04 1.891e-04 1.197 0.265
## GSTA2 1.046e-03 7.294e-04 1.434 0.190
## fwit -1.746e-01 3.801e-01 -0.459 0.658
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4.489 on 8 degrees of freedom
## Multiple R-squared: 0.8179, Adjusted R-squared: 0.6358
## F-statistic: 4.491 on 8 and 8 DF, p-value: 0.0241Figure 23: Generating a model using the original biomarkers and incorporating wit.